EP0034070A2

EP0034070A2 - Fault tolerant memory system

Info

Publication number: EP0034070A2
Application number: EP81300562A
Authority: EP
Inventors: Vernon George Mckenny; David L. Taylor
Original assignee: Mostek Corp
Current assignee: CTU of Delaware Inc
Priority date: 1980-02-12
Filing date: 1981-02-11
Publication date: 1981-08-19
Also published as: EP0034070B1; GB2082005A; GB2082005B; DE3174600D1; WO1981002360A1; US4281398A; JPH0320840B2; CA1163374A; EP0034070A3; JPS57500128A

Abstract

lock redundancy is utilized to improve yield and lower die cost for an electrically programmable read only memory (EPROM). The EPROM is organised 8Kx8 with four primary memory blocks 12A, 12B, 12C, 12D on each side of a central row decoder 14. Each block includes an array M of memory cells, column select CS column decode CD, sense amp SA, data buffer DB and other overhead circuitry. One block of redundant circuitry 12RB-1 is also provided for each set of four blocks and includes a redundant memory matrix RM, a redundant column decoder RCD, a redundant column select RCS, a redundant sense amplifier RSA and a redundant data buffer RDB. Incorporated within each primary memory block is a multiplex logic circuit 26 MUX which is independently programmable to selectively disconnect the associated primary memory block and substitute the redundant memory block, including the redundant column decoder, column select, sense amp and data buffer. Each multiplex logic circuit 26 includes a polysilicon fuse which is permanently programmable from a closed to an open circuit condition by applying a high voltage to the external data bit terminal (P1, P2, ...) which corresponds to the defective memory block.

Description

The present invention relates to a fault tolerant memory system of the type defined in the introductory part of claim 1 and such as may be implemented as a monolithic memory array on a semiconductor chip.
Large scale integration (LSI) techniques have made possible the construction of memory devices having large arrays of binary storage elements on a single chip of silicon. The immediate advantages for such arrangements are the high cell density and low power requirements. In the production of monolithic chips, it is not unusual for the yield of good chips from a silicon wafer to be low, especially during early production runs. For each perfect chip produced, there are a number of chips that are almost perfect, having one or more localized defects which render unusable a single cell or a few closely associated cells or clusters of cells.
It will be appreciated that the presence of only one defective cell in an otherwise perfect memory array can render useless the entire memory array.
As cell density increases, the likelihood of processing defects increases. Therefore there is a continuing interest in techniques for improving the yield of perfect arrays, and for repairing or otherwise rendering usable those memory arrays having processing defects.
Several prior art approaches have been implemented for improving yield. For example, error correction codes have been used to correct words read from a memory in which certain bits of a word are stored in defective cells. According to another approach, a discretionary wiring technique is used during processing to bypass defective cells. Additionally, defect-tolerant memory systems have been disclosed in which an entire redundant row or column of cells is substituted for a selected row or column containing one or more defective cells. In such an arrangement, a redundant row of perfect cells is substituted for a row having one or more defective cells by storing the. word address of the defective row in a content addressable memory along with the address of the redundant row.
In yet another fault tolerant arrangement, a cell addressable array utilizes a redundant row of cells together with a defective word address register and a comparator circuit for disabling a defective row of cells and replacing it with a redundant row of cells. The word address is stored either by selectively open-circuiting conductive paths in a read only memory, or by selectively grounding bits of the read only memory.
In each of the foregoing fault tolerant arrangements, whether the memory be of the word addressable or cell addressable type, a requirement in each is that at least one: redundant line of cells be provided for each row or column in which one or more bad eells exist. Further, faulty bit locations of a memory array can be tolerated only to the extent that rows or columns in which such faulty locations exist do not exceed the total number of redundant rows provided.
Additionally, cell and word addressable arrays typically include decoders, input/output logic, and other overhead circuitry, in which processing defects may occur, and which cannot be cured by word addressable or cell addressable techniques.
Accordingly, the principal object of the present invention is to provide an improved fault tolerant memory array.
Another object of the invention is to provide a fault-tolerant memory system for curing defects in decoders, input/output logic and related overhead circuitry, as well as cell defects.
Still another object of the invention is to provide a fault tolerant memory array in which circuit defects occurring after encapsulation can be cured by programmable means without tear-down of the package.
A further object of the present invention is to provide an improved fault-tolerant memory system in which a memory array having one or more localized defects which render unusable a single cell or a few closely associated cells or clusters of cells can be rendered usable without knowledge of the address locations of the cells.
The fault tolerant memory system according to the invention is characterised by the features set out in the characterising part of claim 1.
According to a preferred embodiment, the programmable means comprises a semiconductor fuse for each primary memory block and fusing circuitry. Upon the detection of one or more defects in a primary memory block, a relatively high voltage, typically 25 volts, is applied to the fusing circuitry through the external bit data pin which causes the semiconductor fuse to be permanently altered from a closed circuit low resistant condition to an open circuit condition, thereby permanently disconnecting the defective primary memory block and substituting in its place the redundant memory block. Each primary memory block preferably includes column select, column decode, sense amp and data buffer circuitry, all of which are duplicated in the redundant memory block, whereby defects in a substantial portion of the cell selection and data detection circuitry can also be cured.
The invention will now be described in more detail, by way of example, with reference to the accompanying drawings, in which:

Figure 1 is a block diagram which illustrates the chip architecture of a memory system having block redundancy and embodying the invention;
Figure 2 is a simplified circuit diagram for the logic multiplexer of the memory system; and
Figure 3 is a simplified block diagram which illustrates the logical interconnection of a primary memory array with a redundant memory array.

In the description which follows, the invention is described in combination with an electrically programmable 8Kx8 read only memory (EPROM) which is implemented by MOS/LSI techniques on a semiconductor chip. It will be appreciated, however, that the block redundant circuitry and technique disclosed herein may be used to good advantage for memory arrays which are organized generally in the form MKxQ.
Referring now. to the drawings, and in particular to Figures 1 and 2, an electrically programmable, read only memory (EPROM) is implemented by MOS/LSI techniques on a semiconductor chip 10. The EPROM is organized 8Kx8 with four primary memory blocks 12 A, B, C, D and 12 E, F, G, H on each side of a central row decoder 14. Each primary memory block includes data storage cells arranged in a memory array M of 256 rows and 32 columns, with the columns of each array M being coupled through data node input/output circuitry for conducting data through common data input/output terminals Pl, P2, P3, P4, P5, P6, P7 and P8, respectively. Each primary memory block also includes column select circuitry 16, column decoder 18, and sense amp and data buffer circuitry 20. Also provided is an address buffer block 22 which serves both row and column decoders, and a common control logic block 24 which sends control signals to all parts of the chip. The row decoder 14, column select 16, column decoder 18 and address buffer block 26 receive the address information and select one out of 256 rows and one out of thirty-two columns for each block to select an individual cell in each block for read/write operations. For example, the row decoder may be of the type which decodes an n-bit binary input signal to generate a one of 2ⁿ signal to select one of 2ⁿ rows, and the column decoder may be of the type which decodes an N-bit binary input signal to generate a one of 2^N signal to select one of 2^N columns in each block.
The cells in each block extend along common row lines, and are separately addressable. The bit column lines in each block are common to two columns of data which are also selectable by two different column addresses. The column bit lines of each block are coupled through data node input/output circuitry (SA & DB) 20 for conducting data to or from the cells of the array through the external pins Pl - P8.
As previously discussed, the principal object of the invention herein is to provide a circuit technique for improving yield and thereby lowering the cost of a semiconductor memory array. This is achieved in the present invention by providing redundant circuitry which can be selectively enabled to replace entire memory blocks having defective cells or clusters of cells, thereby salvaging an otherwise defective chip. As will be seen below, the redundancy technique of the present invention may be carried out equally well during the testing stage both before and after encapsulation.
The redundant technique of the invention is "block" redundancy in that one full block of memory is replaced, rather than one row or one column. In the memory array implemented on the chip 10, data storage cells are arranged in 256 rows and 256 columns, with the columns being grouped in eight outputs, so that the memory array is organized in eight memory blocks which are each thirty-two columns wide. According to the redundancy technique of the invention, two additional blocks, redundant matrix blocks 12RB-1 and 12RB-2 are included as the ninth and tenth blocks of the matrix. Each redundant block includes data storage cells arranged in a redundant memory array RM of 256 rows and 32 columns, redundant column select circuitry (RCS) 16, redundant column decode circuitry (RCD) 18, and redundant sense amp and data buffer circuitry (RSA & DB) 20. The data storage cells in the redundant blocks are addressed simultaneously with the cells of the primary memory blocks.
Upon determining during testing that some portion of a memory block is bad, that memory block is selectively disconnected by programmable means and the redundant memory block for that side of the chip is connected in its place. The redundant block is selectively inserted for the block of the matrix that is bad, not only as probed during wafer testing, but after encapsulation during data retention tests. If a bad bit is detected, the redundant block can be selectively substituted for the defective block either at the wafer level or at the package level. According to this technique, even though only one bit is bad, an entire block of memory is replaced, not just a single row or column. This block redundancy technique is independent of any address in consideration because it duplicates the entire column address and row address sequence.
Block selection is implemented through a multiplexer 26 and a repair buffer 28 which are coupled to the data node input/ output circuitry (SA & DB) 20. The redundant block substitution is carried out by selectively applying a high voltage on the output data terminal of the block in which the bad bit is located. Blowing a polysilicon fuse 30 in the repair buffer (RB) 28 electrically disconnects the bad block of memory while simultaneously substituting the redundant block in its place.
Referring now to Figures 2 and 3, each sense amplifier and data buffer 20 comprises a sense amplifier SA for READ operation and a data buffer DB for write operation connected to the common input/ output node 32 or 34 of the primary or redundant memory block. Considering READ operation (which is all that may be provided for in a ROM), the. sense amplifier has an output node 36 for the primary block or 40 for the redundant block. These two nodes are connected to the two inputs of an output multiplexer OUT MUX corresponding to the primary data block in question. It should be understood that this arrangement is duplicated for each primary memory block and also for the opposite side of the chip 10, and that similar circuitry for both sides of the chip is included for WRITE operations. The READ operation for the memory blocks of the left side of the chip 10 will suffice for explanation of the redundant technique.
The multiplex logic circuit 26 output multiplexer has first and second data inputs connected to the sense amp output node 36 and RSA output node 40 from the primary memory block 12A and redundant block 12RB-1, respectively. The data output of the multiplexer is conducted through an output buffer 38 to pin Pl. The data appearing on either the primary memory block SA output line 36 or on the redundant SA output line 40 will be conducted depending upon the state of the programmable fuse 30 in the repair buffer 28. Assuming that the memory block 12A contains an array of perfect cells, and that the programmable fuse 38 is intact, data will be conducted from the SA output line 36 to the I/O bit data terminal Pl through a common data bit node 42 which forms the input to the output buffer 38.
The condition of the programmable fuse 30 determines the data path selected by the multiplexer 26. Multiplexer 26 (Fig 2) includes first and second input control lines 44, 46 which are coupled to outputs of the repair buffer 28. A logic high level potential, V , or logic low potential, zero or substrate reference potential, will appear on the control input lines 44, 46 depending upon the open circuit or closed circuit condition of the fuse 30.
Assuming that the memory block 12A contains an array of perfect cells, the fuse 30 will be left intact. The resistance of the programmable fuse 30 is relatively low, typically 100-200 ohms, whereby a low resistance conductive path is established for a control transistor Ql. The gate of transistor Ql is coupled directly to V_CC, and is turned on, thereby establishing a conductive path for depletion mode transistor Q2. The depletion mode transistor Q2 is normally on, in the absence of a turn-off signal, and is turned off when Ql begins conducting through the fuse 30. Although the voltage drop across the fuse 30 is not zero, the potential applied to the gate of depletion mode transistor Q2 is substantially at substrate reference level, thereby causing it to turn off.
The potential at the control node 48 is at substrate reference or logic zero level. The gate of a control transistor Q3 is coupled to the control node 48, and is also turned off at the same time. The source of control transistor Q3 is connected to the drain of a depletion mode transistor Q4 through a control node 50. The gate of the depletion mode transistor Q4 is coupled to the control node 50, so that the control node 50 rises to a potential level of approximately V_CC when Q3 turns off. Thus the multiplex control input line 44 rises to V_CC or logic high condition, thereby turning on control transistors Q5 and Q6 in the multiplexer 26, thereby enabling data input and data output from pin PI to the data storage cells of the primary memory block 12A.
If, during testing, it is determined that one or more cells in the memory array M of a primary block is bad, for example block 12A, it is desirable then to substitute the redundant block 12RB-1 in its place. This is carried out by conducting a signal RPR on the gate of a control transistor Q7 while at the same time impressing a relatively large voltage, for example 25 volts, on pin Pl. The programming signal represented by the relatively high potential is conducted through a program input conductor 52 which is coupled to pin Pl and to the source of Q7. As signal RPR rises from logic zero to logic one, the complement repair signal RPR falls to logic zero, thereby turning off control transistor Q8. Programming current is conducted through transistor Q9 when the relatively high voltage carried by the program input conductor 52 is applied to its gate. The programming current is derived from a separate power supply potential Vpp. The application of the repair signal to the gate of Q9 causes it to conduct, thereby delivering heavy current flow through the fuse 30.
After the fuse 30 has been opened by this heavy current flow, the drain of control transistor Ql is open circuited, whereby its source, which is coupled to the drain of depletion mode transistor Q2, rises to V CC potential level. With the potential of node 48 being V_CC, the multiplex control input 46 rises to V_CC, or logic one, and at the same time, transistor Q3 is turned on, thereby driving control node 50 to logic zero potential. Since multiplex control input 44 is coupled to the control node 50, the control transistors Q5 and Q6 for the primary block 12A are turned off, while control transistors Q10 and Qll are turned on by the logic high potential level on multiplex control input 46. Thus with the fuse 30 permanently programmed to its open circuit condition, the redundant block 12RB-1 is permanently substituted for the primary block 12A, whereby data may be written into and read out of the memory cells of the redundant matrix RM through I/O data pin Pl.
It will be appreciated that the foregoing technique allows most memory array defects to be repaired, and in addition, allows the replacement of defective column decoders, column select circuitry, data-in buffers and the sense amp. Although many row line defects cannot be repaired, all individual memory cell problems and column problems can be cured.
Previously, redundancy in a memory chip has been implemented by providing a few additional rows and/or columns. While this approach requires less matrix area, it cannot repair column decoders or sense amps, and in addition has its own inherent implementation problems, including the requirement for a programmed logic array to store cell address information. For a certain memory arrangement, in which a column bit line is shared by two columns of data which are also selectable by two different column addresses, a drain-to-source short or drain-to-floating gate short in the memory cell would be nonrepairable unless the bad column address line is permanently connected to ground at the same time the column address is activated. This would require a polysilicon fuse and considerable additional circuitry to be associated with each of the densely packed column address line. Likewise, the circuit designer must assure that both ends of a replaced row line are permanently connected to ground. This leads to even more severe circuit packing density problems since the row address line pitch is twice as dense as a column address line pitch, and also since there are four ends of the polysilicon row line, as in the present example, when the row decoder is placed in the middle of the memory array. This approach requires a large amount of overhead circuitry.
Since the redundant block approach of the present invention replaces an entire array, including column decoder, sense amp and other overhead circuitry, defects in those areas as well as memory cell defects can be easily replaced. Although block redundancy increases the area of the array, column decoders, sense amps, etc., by 25%, it still requires only the original amount of area to be functionally good. The increase in total chip area is 20%, but the increase in required good active area is less than 2%. The chip area required to implement one redundant column per output in four redundant rows significantly exceeds the area required to implement the block redundancy approach described herein..

Claims

1. A fault tolerant memory system comprising a plurality of primary memory blocks having data storage cells arranged in rows and columns and redundant storage cells for replacing defective cells in a memory block, characterised by a redundant memory block (12RB) of data storage cells arranged in rows and columns, data detection means (20) coupled to each primary (12A-B) and redundant (12RB1), memory block, each data detection means each including a data node (36, 40) for separately conducting data from the storage cells of the block to which it is coupled to an external data terminal (Pl, P2, ...) in response to row and column address signals, a plurality of multiplexer circuits (26) each having a first input connected to the data node (36) of the data detection means (20) of a respective primary block and a second input connected to the data node (40) of the data detection means of the redundant block, and an output coupled to the corresponding external terminal (Pl, P2, ...) and independently programmable - means (28) for each multiplexer circuit (26) having a first stable state in which the first input is connected to the multiplexer output and capable of being actuated to a second stable state in which the second input is connected to the multiplexer output.

2. A fault tolerant memory system according to claim 1, characterised in that each data detection means comprising a sense amplifier (SA) and a data buffer (DB).

3. A fault tolerant memory system according to claim 2, characterised in that each multiplexer circuit (26) also has an input distributor (IN MUX) with an input connected to the corresponding external terminal (Pl, P2, ...) and first and second outputs enabled in the first and second stable states respectively and connected to the corresponding primary block node (36) and the redundant block node (40) respectively.

4. A fault tolerant memory system according to claim 1, 2 or 3, characterised in that each independently programmable means (28) comprising a repair buffer (28) including a polycrystalline silicon fuse (30) and a gate circuit (Q7-Q9) coupled to the fuse for conducting heavy current flow through the fuse in response to a repair signal (RPR) applied to the gate circuit and a programming voltage applied to the corresponding external data pin (Pl, P2, ...).

5. A fault tolerant memory system according to any of claims 1 to 4, characterised in that each multiplexer circuit (26) includes first and second gate circuits (Q5 and Q10) connected in series between the output nodes (36 and 40) of the corresponding primary memory block and redundant block respectively, and the corresponding external data terminal (Pl, P2 ...) and that each independently programmable means (28) includes a fuse (30) and a switching . circuit (Q1-Q4) coupled to the fuse, the switching circuit having first and second output nodes (50 and 48) coupled to the first and second multiplexer gate circuits (Q5 and Q10) respectively, the switching circuit developing enable and complement enable signals on the first and second output nodes respectively, when the fuse is intact and developing complement enable and enable signals on the first and second output nodes respectively, when the fuse is in open circuit condition.

6. A fault tolerant memory system, characterised by the combination of a plurality of primary memory blocks each including a plurality of rows and columns of data storage cells; a redundant memory block including a plurality of rows and columns of data storage cells; cell selection means coupled to each primary and redundant memory block for simultaneously addressing a cell at identical row and-column addresses in each block; data detection means coupled to each cell selection means having a data node separately conducting data to one of a plurality of external data terminals; each multiplexer having first and second control inputs, first and second data inputs, and a common data output, the first and second data inputs being connected to the data node of the primary block data detection means and to the data output node of the redundant block data detection means respectively, and the common data output being coupled to the corresponding primary block external data terminal; and independently programmable means coupled to each multiplexer for logically enabling data transmission exclusively from the corresponding primary memory block or exclusively from the redundant memory block.

7. A memory system of the type including a primary memory block having data storage cells arranged in rows and columns and cell selection/data detection circuitry for conducting data from the cells to an external data terminal, characterised by a redundant memory block having data storage cells arranged in rows and columns and cell selection/data detection circuitry for conducting data from the cells of the redundant blocks; a multiplexer interposed between the primary memory block, the redundant memory block and the external data pin for selectively enabling the transmission of data exclusively from the primary memory block or exclusively from the redundant memory block to the external data terminal; arid independently programmable means coupled to the multiplexer for controlling its operation, said programmable means being characterized by first and second programmable stable states, wherein data transmission from the primary memory block to the external data pin is enabled in response to the first stable state, and data transmission from the redundant memory block being enabled in response to the second stable state.